Ilmu Komputer & AI editorial
Open AccessOA2026
SPFR: Semantic Potential Field Routing for the Distributed Internet of Agents
Hop-by-hop executor discovery and reselection via task-conditioned semantic potentials, with loop-freedom and finite-hop termination guarantees
Yeguang Qin; Liangqi Peng; Fengxiao Tang; Ming Zhao· 2026· DOI 10.48550/arXiv.2608.25396
The core problem
In a distributed Internet of Agents (IoA) without centralized routing control, routing tasks to capability-matched executors is challenging because destinations are not predetermined and agents have bounded local service views. Discover-then-forward approaches select an executor before network forwarding and therefore do not directly support reselection when additional candidates become visible downstream. This creates a structural mismatch: the best executor for a task may only become visible several hops into the forwarding path, yet the decision has already been committed. SPFR addresses this by integrating executor discovery and reselection into hop-by-hop forwarding, allowing the forwarding agent to revise its choice as new candidates enter its local semantic forwarding information base (FIB).
Innovation
Extensive simulations on real-world topologies show that SPFR approaches the realized utility of distributed utility-greedy routing and request-triggered global discovery while using fewer forwarding hops and substantially fewer request-triggered messages. The algorithm remains robust under network and service dynamics. These results indicate that embedding discovery and reselection into forwarding can retain near-optimal utility without the message overhead of global discovery or the rigidity of pre-committed executor selection.
In a distributed Internet of Agents (IoA) without centralized routing control, routing tasks to capability-matched executors is challenging because destinations are not predetermined and agents have bounded local service views. Discover-then-forward approaches select an executor before network forwarding and therefore do not directly support reselection when additional candidates become visible downstream. This creates a structural mismatch: the best executor for a task may only become visible several hops into the forwarding path, yet the decision has already been committed. SPFR addresses this by integrating executor discovery and reselection into hop-by-hop forwarding, allowing the forwarding agent to revise its choice as new candidates enter its local semantic forwarding information base (FIB).
SPFR represents each executor visible in a local semantic FIB as a task-conditioned semantic potential source. The utility of an executor sets the strength of its potential, while hop distance induces exponential attenuation. Formally, for a task and a visible executor at hop distance from the forwarding agent, the potential is:
Why it matters
The theoretical results establish that, under task-consistent frozen-FIB conditions, SPFR is loop-free and terminates in finite hops, with an explicit additive error bound under bounded visibility relative to the full-visibility objective. This bound quantifies the cost of local service views and clarifies when hop-by-hop reselection is sufficient. Practically, SPFR's advantage stems from deferring commitment: because the dominant executor is recomputed at every hop, downstream candidates can displace earlier choices without a separate discovery round. The reduction in request-triggered messages is particularly relevant for distributed IoA deployments where centralized routing control is unavailable. The robustness under network and service dynamics suggests the potential-field formulation degrades gracefully as utilities and visibility change, though the frozen-FIB assumption remains a key condition for the formal guarantees.
Who should read this
CS practitioners and researchers
Opening member content…